CHARACTERIZATION, REPRODUCTION, AND RESOLUTION OF SOLDERJOINT MICROVOIDING

Donald P. Cullen MacDermid, Inc.

Waterbury, CT USA Dpcullen@macdermid.com

(originally published as part of the proceedings of IPC EXPO/APEX ECWC World Conference February 2005

Anaheim CA, USA.)

Abstract Microvoids are tiny voids in solderjoints and differ from more well known solderjoint voiding in their individual size and location. The microvoids discussed herein are described as an abundance of small voids at or near the interface of a PCBA solderjoint. In the most severe cases, a solderjoint my fail physically and electrically. Each void reduces the cross-sectional area of the solderjoint; at some point the remaining solder is insufficient to meet functional demands. While information related to these tiny interfacial voids has existed in industry literature for several years, the use of recently available X-ray analytical equipment has raised the level of microvoid observation. It is not known if microvoiding is responsible for previously failed assemblies to which no root cause failure mode had been assigned. Unlike “Black Pad” interfacial fractures related to Electroless Nickel Immersion Gold, microvoiding has not been directly related to a galvanic effect of PCB circuit design. This provides hope that the phenomenon may be more easily prevented, even on the most difficult designs.

Figure 1: Solderjoint Microvoiding by SEM (top-down) and Optical (cross-sectional)

This project summary is somewhat chronological. First, a description of solderjoint microvoiding is presented. Next, the published literature discovered on the topic is reviewed. A set of experiments was conducted by project teams to screen the most influential variables. A major breakthrough arrived with the laboratory reproduction of microvoids. Once controlled, the primary factors affecting microvoid creation were quantified statistically. Finally, a set of process changes were studied and some recommendations offered. In particular, care should be taken to control copper surface preparation, silver thickness, and reflow temperature metrics. This paper focuses on the impact of immersion silver and how variations in its processing affect voiding. Caution should be taken throughout this summary: only select factors were studied in-depth. For example, the impact of non-silver surface finishes was not quantified. Other possible factors should be investigated to determine and quantify their contributions to microvoid creation. Introduction Solderjoint voiding in assembled electronic equipment is a common phenomenon. The factors identified as causing solderjoint voids include flux type, soldering temperature/dwell, surface cleanliness, and design factors such as microvia structure. Numerous technical articles continue to be published on this topic.1,2 Of note recently is IPC’s 7095 specification describing the location, size, and specification of solderjoint voids.3 Microvoids, also known as champagne voids and champagne bubbles, are a subset of general solderjoint voids. While solderjoint voids vary in size and location, microvoids may be defined as voids less than about 40 micrometers in diameter. Microvoids, like all solderjoint voids, become troublesome when they exist in sufficient number to reduce the overall cross-sectional

area of the physical solderjoint. In the most extreme case, the voids may exist in the thousands per soldered feature. Perhaps most importantly, microvoids may exist in a plane just above the interface with the PCB. In this study, microvoiding was found just above the intermetallic layers formed between the copper from the circuit board and the tin from the solderpaste.

Figures 2a,b: Voiding Size in Solderjoints – Microvoiding and Macrovoiding

Higher quality inspection of solderjoints, together with increased inspection frequency, allows for the detection of microvoiding. One might consider that microvoiding has always existed in solderjoints, but in the past, solderjoint analysis techniques were insufficient to detect microvoiding. In the past, very small voids might have been dismissed as artifacts of cross-sectioning. Newer X-ray inspection equipment has allowed a major advance in inspectability of solderjoint microvoids. In particular, higher energy equipment coupled with fine-resolution detectors permit the discovery of microvoids in a production setting. The use of such equipment allows for the detection of phenomena occurring in low frequency, such as microvoiding. Another factor affecting the increased detection of microvoiding recently is the assembly of higher density circuitry with challenging assembly temperature profiles. Very complex circuit boards are far more difficult to reflow due to temperature variation within the PCBA. Even with great care, temperature variation may exceed 20ºC. Results of NEMI investigations reveal that in order to decrease a 20+ºC temperature delta, engineers required an additional reflow module and decreased line speed. The optimized temperature delta was 16ºC.4 To avoid excessive temperature exposure of sensitive components, low temperature PCB areas may experience insufficient soldering energy. In this scenario, fluxes volatilize, but do not leave the solderjoint sufficiently before the solder hardens, entrapping voids.5 The changes described above occur at a time coincidental with other major changes in the electronics assembly industry. In particular, the trend towards “Lead-Free” assembly has introduced a new set of materials. Lead-free solder itself may have properties affecting microvoid formation due to surface tension and flux choice.1 Higher temperature assembly, associated with Pb-free materials, exacerbates temperature variations during assembly. The Pb-free transition has also introduced new PCB surface finish materials. Immersion silver is experiencing a large increase in use due to its many benefits. With wide deployment of immersion silver, estimated at 5-15% of the PCB market in 2004, a certain number of observations of microvoiding on immersion silver coated PCB’s might be expected. Since the use of immersion silver is coincident with better X-ray inspection capability, a hypothesis formed linking microvoids to immersion silver. As a leading supplier of immersion silver to the PCB manufacturing industry, MacDermid’s R&D sought to investigate this hypothesis. This article focuses on the prevention of microvoids almost exclusively from the perspective of the silver PCB surface finish. Literature Review Several studies exist in published literature with reference to solderjoint microvoids. A 2000 Lucent study of ENIG was the cover story to IEEE’s Transactions. The black-pad study produced unexpected failures from a crack propagating through a string of tiny voids located just above the intermetallic. “The tendency for aggressive immersion Au plating baths to induce voiding is documented by Cordes and Huemoeller6. In their studies, they correlate severe voiding with a reduction in shear strength but they do not correlate this effect with thermal cycling data. There is evidence from other studies that arrays of small void can effectively shorten the crack path, resulting in a moderate reduction in solderjoint fatigue life in a thermal cycling test. Voiding has a subtle effect on the quality of the solder joints and that is reflected in a lower Weibull slope for those affected packages.”7

Figures 3a,b: Optical Views of ENIG Microvoiding in Cross-Sectional and Planar Views7

Voiding, and the effect on reliability at Motorola, is discussed by D.Banks et al. In this study, the presence of a small amount of voids was shown to be of a reliability benefit, as a method to allow relaxation of the solderjoint when presented with physical stress. The propagation of cracks was interrupted when cracks encountered voiding.8 One theory attempting to explain solderjoint microvoids employed the Kirkendall effect. At IEEE’s 2004 ECTC, Texas Instruments reported solderjoint fractures at the intermetallic/copper interface resulting from the formation of tiny voids. Kirkendall voids form when the rate of dissolution of one material (copper) into another (tin) is unbalanced, leading to displacement voids. This seems different from microvoiding in two respects, the voids occur at the copper side of the intermetallic, and the Kirkendall voids appear to grow with post-soldering temperature exposure.9

. . Figure 4: Kirkendall Voiding9 Figure 5: Fracturing due to CTE Mismatch In a 2001 Solectron/ Smart Modular Technologies report “Impact of PCB Surface Pad Finish and Contamination on BGA,” the authors determined that organic contamination of an electrolytic nickel/gold PCB finish produced solderjoint voiding sufficient to cause package separation. The report recommended closer monitoring of the PCB fabrication processes to prevent such contamination.10

Figure 6: Solderjoint Fractures due to Voiding10

Rockwell Collins detected solderjoint microfractures on tin-lead reflowed immersion tin. The microvoiding was investigated in response to a conspicuously reduced Weibull distribution of solderjoint reliability. The 2000 paper, “An Investigation of the Effects of Printed Wiring Board Surface Finish and Conformal Coating for Ball Grid Array Assembly”11 shows the improvement in solderjoint reliability due solely to the better control of immersion tin deposition. Most recently, Dage, a maker of high-powered X-ray inspection equipment, published summaries of void/microvoid inspection testing in June’s edition of Circuits Assembly and at SMTA International 2004. Dage stated that the primary causes of void formation include insufficient reflow temperatures, trapped moisture, and the use of new solderpastes. The new pastes require higher temperatures, employ larger flux volumes, and may have higher surface tensions. OSP was determined to cause the highest number of void occurrences. On microvoids, the author stated; “This voiding phenomenon has been found on PCBs with OSP, ENIG and immersion silver finishes. It is seen infrequently and does not exhibit a defined pattern, which points to a random failure mode, not simply a material mismatch.”12,13

Figure 7: Detection of Microvoids by 2D X-Ray Equipment12

The IPC released a draft of the new BGA reference guide IPC-7095A Design and Assembly Process Implementation for BGAs3 which recognizes the “type E” condition of small voids near the PCB interface. IPC-7095’s proposed maximum for such voids by X-ray inspection is 20% of the solderjoint diameter for Class III product, or 50% of the diameter for Class I equipment. Jisso’s documentation adopts a similar recommendation.

Figure 8: From IPC-7095A Design and Assembly Process Implementation for BGAs3

Initial Investigations The initial investigation of a link between microvoiding and immersion silver was prompted by a high visibility failure of a PCB assembly at post-assembly power cycling. The investigation of the failure received immediate priority. Assigning priority to failure investigation may employ Failure Mode and Effect Analysis (FMEA) methodology.14 The FMEA Risk Priority Number (RPN) is calculated from three key factors: severity, occurrence, and detectability. In this case, the severity of massive microvoiding was electrical failure (high.) The detectability of the microvoiding, on the existing assembly inspection stations, was poor due to the lack of X-ray units. The occurrence of the defect leading to electrical opens was low, resulting in just one or two PCBA’s. The resulting RPN was sufficient to prioritize investigation of the microvoid occurrence on immersion silver boards. A first step in the investigation process involved the formulation of theories surrounding microvoid expression. A list of factors thought to be possible contributors is presented in Table 1. At the beginning, the failure was thought to be entirely related to contamination and residue. Samples with actual functional failure did exhibit severe soldermask residue, which limited the available pad area for good soldering, and led directly to a weakened solderjoint. However, microvoids did appear on X-ray images of other BGA solderjoints that were not overly contaminated with soldermask. More theories relating soldermask to microvoiding emerged. Does soldermask contribute to microvoiding by providing a source of volatilized organics? Or will physical fractures occur only where the solderjoint is already weakened by excessive soldermask residue? The issue of contamination remains as a leading contributor to microvoid vulnerability. Early samples demonstrating microvoids were traced to fabrication process areas with process control issues. Contamination on boards included oils, developer foam, tin resist and films of sublimated soldermask volatiles. Cross-contamination from shared rinses, conveyors for OSP, and nearby immersion tin provided additional residue. The silver process was improved to provide adequate pre-cleaning, silver bath analysis, use of a cleaner and microetch, dedicated rinsing, and improved rework procedures. After these steps were implemented, the X-ray occurrence of microvoiding was nearly eliminated. The next step was to determine if microvoiding existed on boards from multiple fabrication locations. Investigation demonstrated that some level of microvoids could be found on PCB’s from all fabricators studied. Similarly, the site of PCB assembly was thought to be a leading factor, but microvoids were found by X-ray on assemblies made at several EMS locations. Increasingly, it seemed that microvoiding was far more widespread than initially believed. Several hypotheses were proposed relating to the effects of copper surface structure on the production of solderjoint microvoids. The copper may be altered in several different processing areas. The deposition of copper itself may produce rough deposits due to current density variations. Tin stripping and soldermask preparation processing may significantly affect the copper structure prior to surface finish. The final finish process employs a cleaner and microetch, but these relatively mild solutions are not adequate to overcome extremely rough or dirty copper. Another process capable of producing a poor copper structure is silver rework. For various reasons, operators may attempt silver rework by processing the PCB repeatedly through the plating bath. This method, along with other silver stripping methods, may easily compromise the integrity of the copper and silver deposits. Once altered by any of the above processes, copper deposits affect the quality of solderjoints. Similar to the way in which a microvia leads to a solderjoint macrovoid, small entrapped air pockets on a very rough copper surface may produce solderjoint microvoiding. Studies at Universal, IBM, and Boeing detail the effects of copper structure on solderjoint reliability.15-18 Table 1: Initial Theories on Microvoid Causes

• Insufficient temperature during solder reflows • Out-gassing of the organic within the silver deposit • Out-gassing of organic brighteners within the underlying copper deposit

• Incorrect/ impure flux chemistry in the solderpaste • Volatilization of water on the PCB or within the solderpaste • Excess oxides and/or tarnish on the silver deposit • Excess oxides and/or tarnish on the copper before surface finishing • Soldermask contamination on the copper • Tin contamination on the copper • Other contamination, such as machine oils, developer foam, etc… • Incomplete microetch alters surface finish grain structure • Surface tension differences between paste and silver • Silver catalyzes flux degradation, slows volatilization • Raw material and/or manufacturing variations in the silver chemistry • Excessive soldermask on the BGA copper pads • Kirkendall voiding – results from differential rates of metal dissolution • Incomplete rinsing of boards at PCB fabrication • Microscopic ‘spheres’ on the surface of the silver deposit • Cross-contamination of OSP with immersion silver at the fabricator • Immersion metal deposition rate affects underlying copper structure, leading to voids • Aggressive soldermask preparation affects copper structure, leading to voids • Rework of surface finish leads to poor copper/silver structure • Insufficient dissolution of silver into the solderjoint (silver and/or organic embrittlement) • Silver bath decomposition creates tendency to microvoid with bath age • Re-deposition of organic-laden cleaners at the end of their bath-life • Handling-induced contamination before component assembly • Oxidation of solder spheres in the paste, flux viscosity

Recreating Microvoids: Testing with Fabrication Partners In 2003, groups of companies representing the chemical supply, PCB fabrication, contract assembly, and OEM sectors began projects to recreate solderjoint microvoids. Sets of test vehicles were fabricated, silver finished, and assembled using variables suspected of causing microvoids. One of the first DOE’s investigated PCBA’s constructed using three levels of silver thickness, two solderpastes, and baking after silver deposition. With no conclusive response, additional experiments were planned. One such experiment is summarized in Table 2. Boards were produced at three fabrication locations. From earlier observations, high thickness was suspected of impacting microvoiding in the production boards. Therefore, thickness was included in the DOE plan. Post-bake and pre-clean were included as a response to fabricator audits at the time of production defects. Hypothesis Fabrication variation can cause microvoiding Fab Location 4 (3 silver and 1 OSP) Silver Thickness 0.25 and 0.75 microns Post-Bake Yes and No Pre-Clean Standard and ½ Cycle Analytical X-Ray, tensile pull and FM determination Results Only 2 samples resulted in microvoids

High thickness result Possible post-bake interaction

Table 2: DOE Summary

This DOE demonstrated the challenge involved in recreating more than a background level of microvoids. This DOE offered no resolution to many of the hypotheses posed before the experiment. In particular, there was not enough response to verify or dispute the theories on contamination, oxides, cleaning, chemical concentration, and rinsing. The only clear response was an apparent connection between silver thickness and voiding. The next efforts were less ambitious. While the experiment in Table 2 attempted to study several aspects of microvoiding at once, the next experiments simply intended to reproduce microvoids. Separate experimentation with a European fabrication partner tested theories relating the plating rate of immersion silver with plating quality and microvoid observation. Initial work seemed to support the idea of plating rate as controlled by pre-dip acidity and silver bath conditions (temperature and concentration.) This study did lead to more intensive investigation of the use of a neutral pre-dip system as a means for controlling initial rate of silver

deposition, and therefore, augmented thickness control. The study was another confirmation that higher thickness increased the tendency for microvoid occurrence. Hypothesis Silver Plating Rate affects Microvoid Creation Plating Time Normal, High Silver Bath Temperature 0.25, 0.75 microns Post-Bake Yes, No Analytical Ball shear, optical inspection Results Suspected microvoiding at high thickness and high

activity silver plating Table 3: Plating Rate Investigation Summary

A follow-up test studied the intentional organic contamination of soldermask vapors on baked immersion silver deposits. The complicated PCB samples did not survive attempts to solder the samples in prototype laboratory reflow equipment. The difficult testing did not produce meaningful results. Next, MacDermid partnered with a US fabricator to test a theory relating microvoids to silver bath age. The simple test vehicle (described below) was assembled on production assembly apparatus with real component connection, but did not recreate microvoids observable by X-ray. Hypothesis Silver bath age affects microvoid occurrence Silver bath age New, Aged, Old Silver thickness 0.18, 0.38, 0.75 microns Pre-dip type Standard, neutral pH Analytical X-Ray, cross-section, ball shear Results Possible microvoids found by cross-sectioning

No microvoids found by X-Ray Table 4: Bath Age Investigation Summary

As a part of this study, chemical samples were sampled from the production silver bath in order to link microvoids to the concentration of some material or contamination in the working solutions. The samples were analyzed using a battery of chromatographic, titrimetric, and spectrographic techniques. Daily bath samples from one fabricator in particular were collected and analyzed from November 2003 until April 2004. Even though an extensive database was collected, no relationship was ever discovered. Additionally, samples were collected from chemical manufacturing sites to discover if some unintentional manufacturing or raw material variation might explain the intermittent observation of microvoids. Once again, no such relationship was uncovered. Recreating Microvoids: Laboratory Methods At this point in time, the project team expressed frustration at the inability to controllably recreate microvoids observable by high-definition X-ray. The team stepped back to bring attempts to recreate microvoiding into the laboratory. When it became clear that a larger number of variables, and therefore a larger number of test boards, were required for microvoid recreation, the project team created alternative test vehicles (TV’s) for quick testing. Sets of 4-layer and 6-layer TV’s were built to allow for SMT soldering of dummy BGA packages (Figure 9.) Additionally, the TV’s provided for reflow of paste and solder spheres. The part could be inspected for microvoids by optical cross-section or X-ray. Ball-shear testing was conducted on parts assembled with paste and spheres. One initial test yielded promising results. A set of simple solderability coupons with BGA pad footprints were plated with increasing levels of silver, screened with solderpaste, and reflowed. The thickest silver sample demonstrated a set of small voids just above the copper pad (Figure 10.) This was the first time microvoids were observed by cross-section in simple experimentation. With the first successful recreation of microvoids in the lab, the next question arose; what characteristic of very thick silver contributes to microvoids?

Figure 9: Test Vehicles Used To Screen Variables in the Laboratory

Figures 10a,b,c: Thickness Induced Microvoiding by Cross-Section of Reflowed Solderpaste

(from left, 1x, 1.5x, and 3x standard thickness) Organic Content Organic content within the circuit board assembly is a leading suspect when predicting solderjoint microvoiding. Studies presented at the Universal webinar link solderjoint reliability to the copper’s brightener content.15 The choice and content of organic materials within solderpaste flux is well known to influence solderjoint voiding. In fact, solderpastes are specifically formulated for prevention of solderjoint voids under various specific conditions.19 Organic material is present not only in the copper plating of outerlayer conductors; it is also present in several of the PCB surface finish deposits. Several factors led to the suspicion that the immersion silver’s organic content might lead to the microvoiding occurrence. Eckel described organic material in a gold metal deposit as a direct cause of microvoids; this organic content was the result of incorrect process control problems at the fabricator.10 Also, the initial silver-coated production failure demonstrated soldermask residue as a contributor to solderjoint weakness. Further, the effect of a flux /temperature interaction of BGA solderjoint voids is well documented. The question remains; does the carbon present in the immersion silver deposit affect microvoid occurrence? Immersion silver deposits may typically contain a small, almost undetectable content of organic material included for anti-tarnishing, grain-refining, or prevention of dendritic growth. While measurement of tiny amounts of material in ultra thin deposits is difficult, studies by Auger and XPS usually result in an approximation between 0.5% and 2.0% w/w of organic in the deposit. The difficulty in measuring the organic on this nanoscale deposit is due to the extreme thinness of the overall deposit. Investigation commenced in the effort to determine if the small amount of carbon material was sufficient to outgas and form microvoids. In theory, one can estimate the amount of material required to form the gas voids in the solderjoint. The ideal gas law PV=nRT may be used to estimate the amount of volume that might arise from volatilization organics.20 The derivation below calculates that the entire silver deposit would need to form a gaseous vapor in order to create an observable void volume. Clearly, another more massive source of volatilizing material, such as flux, soldermask residue, or PCB contamination must be involved. Thickness of Silver 0.5 micron = 5e-5 cm Area of 0.6mm BGA pad ππππ(0.03cm)2 = 2.83e-3 cm2 Volume of Silver on 0.6mm BGA pad 1.423-7 cm3 Grams of Silver on 0.6mm BGA pad 1.49e-6 grams (at 10.5 g/cc Ag) Grams of Carbon on 0.6mm BGA pad 1.49e-7 grams (if C is 10% of ImmAg w/w%)

P = pressure in atmospheres 1 atm n = number of moles of material = g/molecular weight 1.24e-8 moles as Carbon R = ideal gas constant 0.0821 (L atm/mol °°°°K) T = temperature of solidification of solder = 183°°°°C 456°°°°K V = nRT/P = (1.24e-8) (0.0821) (456°°°°K) / 1atm 4.64e-7 Liters of solderjoint gas volume = 462 µµµµm3 Area of voids 1 micron in diameter on a BGA pad 462 µµµµm2 Area of 0.6mm (600µµµµm) BGA pad = ππππ(300µµµµm)2 282,743 µµµµm2 Ratio of void area to pad area 462/282,743 = 0.16% (less than 1% of BGA area) Even if the silver deposit is 10% carbon and all organic vaporizes without escaping the solderjoint, microvoiding is < 1% of BGA area. The assumptions in this model act to overestimate the predicted void volume. The typical thickness of silver is less than 0.5 micron. The organic content is less than 10%. ‘Real’ gasses typically occupy less volume than ‘ideal’ gasses. More direct investigation of the organic out-gassing theory was conducted through the use of Thermal Gravimetric Analysis and Gas Chromatography of heated samples. In both sets of experiments, any significant outgassing from the silver deposit would have been detected as a weight loss or a chromatographic peak. No outgassing was detected. As contrast, when a silver-plated sample was contaminated with undeveloped soldermask residue, large peaks were observed on GC/HS when heated to 215ºC.21 To determine if the silver metal itself may have been causing an interaction, electrolytic silver samples were prepared. A pure deposit of electrolytic silver was deposited on laboratory test vehicles using a cyanide system. Microvoids were found in small quantity by cross-section but not by X-ray inspection, so the impact of silver purity on microvoiding remains unclear. Other tests were conducted in the attempt to determine if any particular ingredient of the immersion silver formulation affected microvoid occurrence. No significance was found as a dependence on any one, or group, of chemicals. Details of these experiments will be published separately. Another characteristic of severe microvoiding was the location of small voids formed into a plane just above the junction of solder with the PCB’s copper surface. If the silver or organic was responsible for embrittling this area of the solderjoint, it was speculated that the silver should remain restricted to this area of microvoid observation. A sample of solderjoint voiding was submitted for elemental mapping. In this experiment and in several other confirming experiments, the silver was found to be completely dispersed throughout the solderjoint. With the completion of each test, more questions were raised. If the thick silver is completely dispersed, how does it interact to affect voiding?

Figure 11: Elemental Mapping Shows No Localization of Silver at the Voiding Boundary

Other theories were proposed to explain the observation of microvoiding at high silver thicknesses. One theory proposed that there may be an interaction between silver and flux causing the flux to more slowly exit the solderjoint. Slowing the flux volatilization/evacuation from the solderjoint should produce increased voiding. To study flux-silver interactions, the materials were heated to 200ºC, evolving gasses were trapped in “headspace” apparatus, and the gasses were quantified with Gas Chromatography. The solderpaste used was identical to the paste used in the initial assembly failures. A baseline measurement was created with volatiles evolved from paste alone. Additional

measurements were acquired with paste screened on silver, OSP, and ENIG surfaces. ENIG and OSP measurements were very similar to the paste-only baseline. The volatiles were significantly lower for the paste-silver samples (see Figure 12.) In this phase of the experiment, a clear interaction was found between the thick silver deposits and the flux/paste material. Somehow, the silver appeared to slow the formation of flux volatiles. One explanation proposed that a catalysis or decomposition of the flux by the silver was enough to slow volatilization. In practice, slowed flux evolution would result in voiding. If a flux interaction occurred at the surface, the voiding might be trapped at/near that surface, as seen in the microvoiding samples. Slower volatilization should exhibit itself in a manner similar to insufficient heat transfer during reflow – locked voids in the solderjoint. Next solderpastes and flux media were studied. An interaction with silver was reproduced with other fluxes, but to a somewhat lesser degree. Microvoids were produced on a variety of solderpastes tested under stressing conditions; of note, Pb-free solderpastes did not produce microvoids in this test.

Volatiles Evolved from Paste on Various Finishes

0 10 20 30 40 50 60 70

flux on ENIG

flux on OSP

flux on Ag

flux only

Gas Chromatograph response (pA)

Figure 12: Interactions between Solderpaste Flux and Surface Finish by Headspace GC

With a better understanding of microvoids in the laboratory, other screening tests continued with fabrication partners. A study was commissioned to measure any impact of poor rinsing on microvoid formation. The hypothesis in this study proposed that residue on the PCB should interfere with good intermetallic formation and may cause outgassing when the residues were soldered. Once again, a thickness range of silver was investigated as well as a storage-time variable between soldermask and silver plating. The storage time may affect the lock-in of residues from incomplete soldermask developing. The residues from incomplete rinsing (short, low pressure, cold rinses) resulted in 6x the ionic contamination compared to TV’s with good rinsing. X-Ray analysis did not show widespread voiding; some isolated pad areas on high-thickness samples and poor rinsing did exhibit microvoiding. Hypothesis Rinsing and soldermask residues cause microvoiding Post-silver rinsing Standard, ½x, 2x rinsing conditions

2-12 µg/in2 (as NaCl) Silver thickness 0.25, 0.50, 0.75 microns Storage time between soldermask and silver 24 hours, 3 weeks Analytical Ionic contamination, solderability, X-Ray, cross-section Results Microvoids found by X-Ray on some BGA pads:

Ag > 0.75 microns and poor rinsing Table 5: Ionic Contamination Summary

Another phase of the investigation soon yielded an additional factor in reproducing microvoids. In further experiments aimed at creating PCB contamination, tin stripping and soldermask preparation processes were investigated. Nitric and fluoboric strippers were used in attempts to incompletely strip tin plating before soldermask. In previous recreations of microvoiding, only very high thickness of silver (0.75 microns) reproduced the defect. In this experiment, a more moderate thickness of 0.4 microns showed microvoiding by cross-section and X-ray (see Figure 13.) That sample was processed in the laboratory with nitric acid based tin stripping. Laser profilometry showed that the surface demonstrated far more roughness than control samples. One explanation proposed that a rougher surface of plated silver should contain the same volume of silver as a smoother surface at higher silver thickness. Another important observation from this experimental phase was that reproduction of microvoiding in

the laboratory required decreased temperatures during solder reflow. In these and subsequent experiments, the peak temperature of the reflow profile was decreased by 10 or 20 degrees centigrade to result in microvoids. Several follow-up experiments gave supporting microvoid observations, but detections were still sporadic. Observations by cross-sectional techniques did not always match X-ray inspection results.

Figure 13: X-Ray Microvoids on Rough Surfaces. Note the Ellipse of Voids at the PCB Surface.

“Turning On, Turning Off” Microvoids The data and conclusions from all the experimental phases pointed toward the interaction of several variables as being necessary to reproduce microvoids. Intuitively, this seemed correct, since a single variable factor would have been much easier to discover, and should have produced a higher frequency of microvoiding on the millions of production parts using immersion silver used in the world. The next experiment intended to combine the factors identified with causing microvoids. At the time of the experiment, the factors suspected were flux type, high silver thickness, rough underlying copper, and cool reflow soldering. The experiment was dubbed the “Perfect Storm” as it brought together the extremes of seemingly unrelated conditions. Hypothesis Several factors set at extreme conditions will allow

microvoids to be “turned on and turned off” Reflow temperature Standard, 20ºC below standard peak Silver thickness 0.38, 1.50 microns Copper preparation Standard, roughened pre-treat and no microetch Analytical X-Ray, cross-section Results (observed microvoids, scale from 1-10)

Average = 7.1, contributing factors Average = 1.1, non-contributing factors

Table 6: Investigation Aimed at Aligning All Known Microvoid Factors The “Perfect Storm” experiment was a huge step forward in the on-demand recreation of microvoids. When a set of 15 test-vehicles was studied, the average microvoid score on a scale from 1-10 (10 denotes severe microvoiding) was 7.1. Comparatively, the control levels of the contributing factors produced an average score of 1.1. Some microvoiding was observed on the control group, particularly in one sample. This observation signified that not all contributing factors were entirely understood. In this experiment, however, the team achieved excellent correlation between the X-ray and cross-sectional methods for detecting microvoids.

Figure 14: Factors Promoting Microvoids Figure 15: Factors Preventing Microvoids

With the increased ability to reproduce microvoids, based on better control of void-inducing factors, the project team moved into a phase of statistical data evaluation. At this stage, there were three goals for performing statistical analysis.

• Ranking the leading factors affecting microvoids • Identifying significance of secondary factors (copper roughness, ionics,…) • Proving reliability of preventing microvoids with process control

In a series of experiments, large numbers of replicates and test conditions were studied using the SMT test vehicle. The results were interpreted using Design Expert software (see Figure 17.) In each test, the results indicated the leading factors to be:

• Cool Reflow Temperature • High Silver Thickness • An interaction between reflow temperature and silver thickness

Other factors did prove some statistical significance in data analysis, but were less reproducible. These factors included the type of copper pre-treatment, ionic cleanliness, and pH of the process chemistry. The effect of copper pre-treatment remained unclear. Some results indicated that insufficient removal of tin and/or soldermask was a leading effect. Other data showed that very rough surfaces, above a Surface Area Ratio (SAR) of 0.05, promoted microvoiding.

R Surface Area Ratio

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07 Rough Texture associated with Microvoid Detection

Surf

ace

Are

a R

atio

Figure 16: Profilometry Measurement of a Microvoiding Sample Shows Extreme Surface Roughness

In most testing, it should be noted that extreme conditions were needed to recreate microvoiding on demand. For example, soldering samples with a reflow profile at 20ºC below recommended peak was the most effective way to reproduce voiding. Similarly, silver thicknesses of 1.5 microns (60 microinches) were needed to achieve microvoiding in the laboratory, while silver deposits on actual production product would not normally exceed 0.60 microns (24 microinches.) It is unclear what might occur when other surface finishes were processed at such extremes. As stated earlier, these studies focused on interactions with immersion silver.

A: Cu Treatment = AverageB: Tin Pre Treatment = Average

E: T ime in Si lver

Sqrt(V

oids b

y X-S

ec)

Std -20

0.000

0.750

1.500

2.250

3.000

1.50 µm Ag

0.40 µm Ag

Std. Reflow Cool Reflow

Voids = 6

Voids = 1

Figure 17: The Effects of Ag Thickness and Reflow Temp (x-axis) on Microvoiding (square root, y-axis)

Assembly Reflow Temperature The assembly interactions related to microvoid creation are strong. Cool reflow profiles showed microvoiding when other factors were outside normal operating conditions. The cool reflow profile used in these studies was achieved by lowering the peak temperature by 10ºC or 20ºC. See Figure 18 for details. While a drastic change such as this was needed to produce microvoiding on simple test vehicles, much more complicated PCB-component assemblies are more sensitive to smaller temperature deltas. In actual assembly, a 20ºC difference in peak temperature may have solderability and non-wetting consequences, but real assemblies are exposed to these temperature deltas.4,22 A study of PCB’s similar to the initial failure discussed above showed that the leading method to create microvoids (on the complex actual production assemblies) was through cooler reflow profiles. Another indication of the importance of temperature is the distribution of microvoiding on BGA footprints. The microvoiding is typically revealed on the periphery of solderballs where cooling occurs fastest (see Figure 14.) Additionally, where surface copper traces provide solderjoints a heatsink, the voids occur in highest density.

Figure 18: Assembly Profiles used in Laboratory Testing Strength of Microvoid Samples Much focus of the above experiments has been on the detection of microvoids by optical microscopic inspection and certain high-resolution X-ray equipment. Not as much investigation in this project team was focused on the reliability of solderjoints exhibiting severe microvoiding. To compare inspection techniques with physical testing methods, a severely microvoided sample was tested for ball shear strength as compared to controls exhibiting no microvoids. Solder spheres were soldered onto a 400 I/O BGA footprint using the factors identified to produce microvoiding. The resulting parts measured 9 on the 1-10 microvoid scale by X-ray. A control part had a 0 measurement. All spheres were sheared using a Dage 4000. The “9” sample showed a significant (30%) reduction in ball shear strength compared to the “0” control, with an even more significant drop (54%) in the mean minus 3 standard deviations (see Figure 19.) Perhaps surprisingly, the severely microvoided sample did not easily fall off the BGA pads.

Ball Shear Strength of Microvoid Solderjoints

0 200 400 600 800 1000 1200

no microvoids

severemicrovoids

gramsmean-3SD mean

Figure 19: Strength of Microvoided Solderjoints Preventing Microvoids While some factors have shown statistical validity as influencing microvoid formation, other factors remain as suspected contributors. Still other factors have not been studied in depth, such as the relative impact of non-silver surface finishes when processed under extreme conditions. For the factors uncovered in this project, there exist ways to prevent microvoid occurrence. Silver Thickness Control: In most experiments, very high thicknesses of silver were required to cause problematic levels of microvoiding. Drastic reduction of immersion silver thickness is not a real option since thicknesses below 0.13 microns may lead to increased tarnishing, migration of copper through pores, and possible loss in contact functions. However, thickness may be reduced as much as practical while maintaining this lower limit. Lower thickness is accomplished in several ways. Primary importance is given to correct measurement of the silver thickness; see IPC-4553 for measurement suggestions.23 Thickness variation may be lowered by ensuring a uniformly microetched surface. Finally, the chemical suppliers will provide ways to control the chemical bath through pH, temperature, and pre-dip to allow optimum rate control. Of course, dwell time in the silver bath provides most direct control, as long as sufficient time is maintained in the cleaner, microetch, and rinses. Deposit Quality: The quality of the immersion silver bath itself has no meaning if other nearby processes are not well engineered and maintained. The copper must be free of organic and tin residues. The roughness of the incoming copper must not be highly variable. The microetch immediately before silver deposition needs to render the copper into a clean, uniform surface. Rework should be prevented, as improper rework may lead to excessive roughness and thickness. If unavoidable, rework must be conducted in strict accordance with supplier recommendations. Silver surfaces with excessive surface area roughness will mimic extremely thick silver deposits. After silver, parts need to be well rinsed, handled and packaged with care, as data shows a link between ionic residue and microvoids. For quality acceptance, tarnish should be prevented, even though no link between tarnish and voiding was discovered. Assembly Control: Action at the assembly house will lead to less vulnerability to voiding, even within maximum temperature restraints. Studies show that adjusting various aspects of the profile, such as pre-heat, ramp rate and peak temperature all affect

microvoiding. Bruno, etal minimizes voiding with medium pre-heat and low ramp rate.5 Use of low-voiding fluxes and better control of paste storage, etc. may impact voiding as well.1,19 However, it is not clear that reducing macrovoids will also reduce microvoiding. With the relatively recent interest in microvoiding, there is limited information linking microvoid prevention to solderjoint voiding in general. More direct study of all assembly-related factors is needed. Conclusions and Recommendations Time has shown that solderjoint fracture due to microvoiding is not a commonly observed failure mode in the industry. The observation of microvoiding by close X-ray inspection reveals microvoiding as a somewhat common phenomenon, but since the use of high-resolution X-ray is relatively new for this exact type of examination, it seems clear that some level of microvoiding has always been present. Microvoid creation is not due to a sole processing factor. Had it been, the problem would have been much more easily detected and resolved. Microvoiding is the result of several interacting conditions. Fixing just one of the interacting factors allows for the control of microvoiding. More investigation is needed into the connection between voiding observation and functional performance by industry-accepted reliability testing. Microvoiding does seem to be associated with the use of immersion silver, and to some unknown extent, the use of ENIG, OSP, immersion tin, HASL and other finishes. Studies show that operating the silver process under recommended process and thickness controls will help prevent the occurrence of functional failures by solderjoint microvoids. Numerous studies, measuring X-ray microvoiding, cross-sections and ball-shear strength, show that high-thickness silver soldered at low-temperature conditions produces microvoids. Thickness control is a very important factor, so methods for better controlling thickness are recommended. Better measurement techniques and sufficient measurement frequency are important. New process changes, intended to provide a more consistent surface texture and silver thickness/volume should be implemented. A quality inspection program, employing thickness measurement and cross-sectional investigations should be installed. In addition to process changes made within immersion silver and other final finish processes, assembly and design factors should be optimized towards the minimization of microvoids. References 1. H.Jo, B.Nieman, N.Lee; “Voiding Of Lead-Free Soldering At Microvia,” SMTA International, Chicago,

September 2004. 2. A.Primavera, R.Sturm, S.Prasad, K.Srihari, “Factors That Affect Void Formation in BGA Assembly”,

Proceedings of Institute for Interconnecting and Packaging Electronic Circuits, October 1998. 3. IPC – Association Connecting Electronic Industries, IPC-7095, Design and Assembly Process Implementation

for BGA’s, August 2000. 4. M.Kelly, Q.Chu, J.Bath, Pb-Free Reflow and Rework, NEMI, Circuits Assembly Nov. 2004 5. F.Bruno, F.Grano, “Effect Of The Reflow Profile On BGA Voiding In A Lead-Free Process,” SMTA Chicago

September 2004 6. F.Cordes and R.Huemoeller, “Electroless Nickel-Gold: Is There a Future?” Future Circuits International, Issue

4, 1998. 7. R.Coyle, G.Wenger, D.Hodges, A.Mawer, D.Cullen, P.Solan, The Effect of Variations in Nickel/Gold Surface

Finish on the Assembly Quality and Attachment Reliability of a Plastic Ball Grid Array, IEEE Transactions on Component and Packaging Technologies, Dec.2003.

8. D.Banks, T.Burnette, Y.Cho, W.deMarco, A.Mawer, “The Effects of Solder Joint Voiding on Plastic Ball Grid Array Reliability,” Surface Mount International 1996, Sept. 1996.

9. C.Chiu, K.Zeng, R.Stierman, D.Edwards, K.Ano, Effect of Thermal Aging on Board Level Drop Reliability for Pb-free BGA Packages, IEEE ECTC 2004.

10. S.Eckel, N.Kini, D.Le, R.Jay, “Impact of PCB Surface Pad Finish and Contamination on BGA Solder Joint Voiding,” SMTA International, Sept. 2001.

11. D.Hillman, N.Cavanah, J.Reinig, L.Keehner, E.Hofer; “An Investigation Of The Effects Of Printed Wiring Board Surface Finish and Conformal Coating For Ball Grid Array Assembly,” APEX 2000

12. D.Bernard, K.Bryant; “Does PCB Pad Finish Affect Voiding Levels in Lead-Free Assemblies?” SMTA International, Chicago, September 2004

13. K.Bryant; Investigating Voids: Does A Connection Exist Between Pad Finish And Voiding In Lead-Free Assemblies?” Circuits Assembly June 2004

14. Failure Mode and Effect Analysis Application Guide, Technicomp Publishing, Inc. 1990. 15. W.Jones, Universal Instruments SMT Laboratory: “So What's in Cu-Pads, and How Do We See It?” Universal

Webinar, September 2004.

16. D.Henderson; “Kirkendall Voiding and Other Embrittlement of Cu-Pads.” Universal Webinar, September 2004.

17. M.Anselm, Universal Instruments, SMT Laboratory: “Embrittlement of Ni/SnPb/Cu Joints,” Universal Webinar, September 2004.

18. T.Woodrow; “High Vibration Testing of Solder Joints,” Universal Webinar, September 2004. 19. Alpha Solderpaste 709; http://www.alphametals.com/whats_new/article.asp?id=133 20. P.Atkins; Physical Chemistry 3rd Edition, 1986 W.H.Freeman & Co. 21. D.Angelone; “Soldermask Contamination of Copper Foil,” MacDermid R&D Internal Report, November 2003. 22. P.Roubaud, J.Gleason, C.Reynolds, K.Lyjak, M.Kelly, J.Bath, “Development of Baseline Lead-free Rework

and Assembly Processes for Large Printed Circuit Assemblies.” 23. IPC – Association Connecting Electronic Industries, IPC-4553, Specification for Immersion Silver Plating for

Printed Circuit Boards (final draft,) October 2004. 24. C.Gurumurthy, “No-Pb soldering to Ni-pads,” Universal Webinar, September 2004. 25. E.Cotts, Binghamton University: ‘Intermetallic Structures in Pb-Free Joints. Solder pad Degradation and Failure

in SnPb Joints on ENIG and Electrolytic Ni/Au,” Universal Webinar, September 2004. 26. P.Kondos, Universal Instruments, SMT Laboratory: “Enhanced 'Black Pad' Sensitivity with SAC Soldering,”

Universal Webinar, September 2004. 27. M.Wickham; Voiding, Occurrence and Reliability Issues with Lead-Free, NPL 28. T.Gregorich, “Brittle Failure of SAC on Electrolytic Ni/Au,” Universal Webinar, September 2004. 29. R.Coyle, A.Holliday, P.Mescher, P.Solan, S.Gahr, H.Cyker , J.Dorey, T.Ejim, “The Influence of Nickel/Gold

Surface Finish on the Assembly Quality and Long Term Reliability of Thermally Enhanced BGA Packages,” IEEE International Electronics Manufacturing Symposium, Austin TX, Oct.1999.